Technical feasibility of egg ingredient reduction/replacement in US foods

Executive summary

1. One potential strategy for reducing animal product usage is to decrease processed egg use in manufactured foods, which represent approximately 8% of global egg production.
2. This report analyzes the US-manufactured food landscape to develop methods and identify opportunities for egg replacement advocacy, examining three key factors:
   1. Volume of eggs used across food product categories
   2. Functions served by eggs in different food products
   3. Technical feasibility of replacing eggs with available alternatives
3. We analyze data from the United States Department of Agriculture’s (USDA) FoodData Central database and Weekly Retail Sales database to estimate egg usage patterns across food categories, and classify egg and replacement functions by food category using data from American Egg Board, Lever Foundation, and Reading Scientific Services Ltd. (RSSL).
4. Key findings from our analysis:
   1. The most prevalent egg functions across food categories are moisture/fat contribution, binding, and structure (medium confidence).
   2. Most food categories rely on eggs for multiple functions simultaneously, making replacement more technically challenging (high confidence).
   3. Categories like prepared subs and sandwiches; prepared wraps and burritos; snack, energy, and granola bars; and pasta may present the best balance between moderate ease of replacement and volume impact (low confidence).
      1. Our preferred robust specification discounts total category egg use by the percentage of egg-containing items and weights primary and secondary egg functions. This approach gives a more conservative estimate of total category egg use and compresses replaceability prioritization, but it largely does not change our overall recommendations (see Robustness checks section).
      2. Notable limitations appear in the selected results (for example, different character iterations of the phrase “cookies and biscuits” are prioritized differently), which highlight the need for more data cleaning research.
5. Recommendations for how to best support egg replacement advocacy:
   1. Conduct further desk research on the top Priority 3 categories: prepared subs and sandwiches; prepared wraps and burritos; snack, energy, and granola bars; and pasta.
   2. Support food and consumer product consultancies who can address aspects of replaceability that go beyond technical feasibility—including costs, manufacturer awareness, and consumer acceptance—for prioritized categories.
   3. Commission scientific research on areas of egg functionality that need more innovation, such as moisture/fat contribution.

Introduction

About 8% of global egg production is processed for inclusion in manufactured foods (Kumar, 2024). Many plant-based alternatives to processed eggs exist, and replacing or reducing eggs with these alternatives could reduce the number of chickens in commercial egg production systems. With increases in egg prices expected to last at least through 2025, food producers may be more open than ever to considering alternatives.

However, what proportion of processed egg use are producers actually incentivized to replace, given the many barriers that have prevented switching so far, like a lack of centralized replacement information (confidential report, Specht, 2025)? Here, we categorize the volume of processed egg use in the US into eight bins, from most to least replaceable. This should help stakeholders determine the extent to which they should try to displace processed egg use now, versus investing in longer-term solutions that will eventually make replacement easier.

The methodology and its limitations are described in detail in the Methods section. For now, we note three important limitations imposed by time and data availability. First, our analysis focuses on entire food product categories, as total sales of individual products are not available. Fortunately, discussion with several food science experts indicates that egg functionality likely does not vary meaningfully across products within a somewhat narrow category (Reading Scientific Services Ltd. [RSSL], Liz Specht; personal conversation, 2025).

Second, we assessed only the technical replaceability of eggs. That is, we ignore potentially important supply-side factors like costs of replacements, and downstream barriers like consumer preference for a clean label (RSSL, 2024). We operationalize technical replaceability along two dimensions: How many functions eggs play in a product category, and how many plant-based alternatives are already commercially available. While there are at least 20 functional roles that eggs can play, no single replacer on its own fulfills all of those functions. Thus, when eggs are simultaneously playing several roles in a product, it is less likely that any single plant-based alternative will suffice.^[1] However, the larger the number of replacements that are already commercially available, the more likely it is that at least one of them will adequately fulfill all or many relevant functions, while also meeting other constraints that are beyond the current scope of investigation.^[2]

Third, this minimum viable product (MVP) focuses on branded food products sold in the US between the years 2019 and 2025. However, we expect the findings to generalize to countries with similar retail offerings to the US. Moreover, the large processed egg use in the US (~26% of 108 billion total eggs) makes it a high-priority region for producer engagement.

Key findings

Table 1 shows total egg use and replaceability prioritization for narrowly defined subcategories (from United States Department of Agriculture [USDA] branded food products data) using our original egg use estimation specification. For the preferred, robust specification results (from which we derive our recommendations), see the Robustness checks section. For ease of visualization, Figure 1 displays the same information using the coarser product categorization of the USDA retail sales data. See Table 9 to understand our criteria for prioritizing replaceability. Note that we intend to update these figures (egg use values and percentages) when we receive the results on egg functionality that we have commissioned from Reading Scientific Services Ltd. (RSSL).

• No product category is classified as Priority 1, in the sense of technical feasibility that we use: the egg function complexity is low (0–0.75^[3]), and there are at least five alternatives to the egg function.
• A large majority of products (~80%) in our sample use processed eggs in a way that may be “moderately” difficult to replace (Priority 3 and 5), meaning that egg function complexity is moderate (0.75–2), and the number of alternatives ranges widely but decreases with decreasing priority. Notable limitations appear in the selected results (for example, different iterations of the phrase “cookies and biscuits” are prioritized differently), which highlight the need for more data cleaning research.
• Roughly 20% of processed eggs in our sample may be “difficult” to replace (Priority 6 and 8), in the sense that eggs play a complex combination of functions in products.

Figure 1: Egg use distribution by priority group

Note: These estimates are low-medium confidence. Percentages are rounded to two decimal places. See Table 1 for disaggregated category results and examples of products in each containing eggs, especially unexpected broad categories like fruit and vegetables (where egg use often occurs in food products containing a majority of fruits or vegetables as well as eggs).

Table 1: Main results from top 10 categories in each Priority level

Note: This table presents egg use estimates and example products for the top 10 food categories in each priority level. Egg use estimates are calculated as the product of per-serving egg use and weekly unit sales, averaged over food categories. Total and percentage egg use numbers are rounded to two decimal places. Table cell colors distinguish priority categories. A known potential upward bias occurs when foods include non-egg ingredients that still contain cholesterol (such as meat, dairy milk, cheese, etc). Full results can be found at https://osf.io/pvmzq/. Due to editorial formatting, some categories may appear to be duplicates although they are distinct in the data. These categories illustrate the importance of additional data cleaning and category investigation before taking strategic action.

Recommendations

• Conduct additional research into the top recommended categories. We recommend targeting prepared subs and sandwiches; prepared wraps and burritos; snack, energy, and granola bars; and pasta for additional research, which then may be appropriate for initial advocacy targets. Additional research is needed for two main reasons: 1) to perform additional data cleaning on the most promising categories to make the cholesterol-based egg estimation method more rigorous, whether through combining iterations of category names or ensuring precision of the ingredient-identifying algorithm, and 2) to further determine trends in egg ingredient types (i.e., whole egg, egg white, or intermediate products like mayonnaise) or major company/brand players. With additional research, initial replacement targets may be identified. Products will likely have high total egg use, relatively fewer egg functions or numerous replacements, and—importantly for initial advocacy targets—consumers may not care about replacing eggs in these products because eggs don’t appear to play an important cultural or nutritional role that is salient when people decide to purchase the products.
  • Recommendation: Perform additional data cleaning on the top four Priority 3 categories, namely to improve:
    • the product category matching process when combining the branded product and sales data, and
    • the egg-identifying string search to be more robust to instances of “eggplant” and “veggies,” special characters, and primarily egg-based products made of whole eggs.^[4]
  • Recommendation: After data cleaning Priority 3 categories, conduct trend analysis on ingredients list and brand or company names to determine 1) common egg types or intermediate egg products, and 2) frequently appearing companies or brands to target.
  • Recommendation: For all Priority categories, many non-technical aspects of replaceability are not easily addressed by desk research alone. Thus, we recommend supporting an egg reformulation consultancy to overcome hesitancy to replace eggs (see related calls for third-party organizations in Specht ,confidential report, 2025, and Grilo ,2024). For example, a consultancy could provide bespoke market research to clients about:
    • Price: Many egg replacers are more costly than eggs and therefore may increase product costs, thereby presenting a barrier to companies, even if replacing eggs is relatively easy. It is not always clear how quickly production could be scaled up if a large company wanted to replace eggs at scale. The higher, more volatile price of eggs may be an incentive to explore replacers, but larger companies need to source a large number of eggs no matter what, and may already be negotiating solutions to mitigate the harm (confidential report, Specht, 2025).
    • Incidental functions: Beyond the core function(s) that explain why eggs are included in the recipes in the first place, eggs may also have incidental functions. Importantly, egg function classification systems do not appear to distinguish between intentional and incidental functions. It seems likely that consumers care less about incidental functions than core functions. This unmeasured difference in function may cause some categories to look more difficult to replace than they may actually be (depending on whether some of their more numerous functions are incidental and easily ignored). A first step toward measuring the difficulty of functions was to include the primary and secondary functionality results from RSSL into the replaceability classification algorithm. However, more comprehensive measures of difficulty and direct vs. incidental will allow for more precise prioritization.
    • Consumer acceptance: Producers may also believe consumers value eggs for reasons that do not have to do with functionality per se (e.g., eggs are consistent with a “clean label” (A. Rondoni, 2021), more consistent with tradition, etc.). Also, it is unclear whether a better marketing strategy would be to quietly drop eggs from the ingredient list, or to promote it as a benefit. And if actively promoting it, should the pitch be “allergen-free” or something else? Does the benefit from reducing egg allergens outweigh the harms from introducing new potential allergens through the replacements (for example, introducing gluten)?
• Robustness checks of within-category egg function. Sample sub-categories within each high-priority category to ensure that functionality does not vary so much as to undermine a single replacement strategy.
• Target specific egg or replacement functionality for future strategic action or bench research. Based on the number of available replacements for particular egg functions, some functions may provide good further research targets. For example, the Binding functionality is both a prevalent egg function across many food categories and has numerous available alternatives, and may not benefit as much from development of additional alternatives as functions like moisture/fat contribution and structure. These both appear frequently in food products but seem to have fewer well-developed alternatives.
  • Recommendation: commission further research into alternatives aimed at replacing moisture/fat contribution and structure functions.

Results

Egg use in food product categories

Our egg use analysis model estimates the number of large eggs that are likely used in the ~1.7 million egg-containing products in the USDA database, then calculates the average egg use in each food product category. These results are then matched to food category-level sales data to understand the market impact of egg ingredients. The results of the ingredient estimation procedure on the package weight subsample can be found in the supplementary file with the path egg_use > results > egg_food_2023_joined_retail_food_data.csv (located in the associated OSF project, along with all other results, data, and analysis code), and Table 2 provides a snapshot of the final results for three randomly selected products.

Table 2: Selected results from the cholesterol-based estimation procedure

Note: the table uses real data and estimation results from a non-validated estimation model. Egg weights and egg-containing product numbers are rounded to two decimal places.

Figure 2 shows the average large egg count of the food categories for products containing egg ingredients. We can see that products with high cholesterol, for example, in the milk and cheese categories, are estimated to be the highest egg users. The assumption that egg is the only contributor of cholesterol likely causes an upward bias in these results. For example, the cow milk in eggnog products (a large egg-containing component of the milk category illustrated in Figure 2) contains cholesterol as well as the whole eggs. We indicate categories potentially affected by this bias in the key findings in Table 1. Nevertheless, according to food science and statistical estimation experts, the cholesterol method involves the least harmful assumptions and methodological limitations for the available data relative to other methods of estimating ingredient use (personal conversation with Specht, IAMECON researchers 2025). For that reason, we choose to use this method to generate point estimates that can be combined with sales data to guide our recommendations and direct further research. While some outlier per-serving results may be overestimated by two orders of magnitude, these outliers seem to be rare, with category average of per-serving for all categories in the single digits (as roughly expected).

Figure 2: Average per-serving egg use in top 20 food categories

Note: Top categories in terms of average egg use. Category names are taken directly from the data and do not have the same editorial formatting applied as the tables in this report.

Adding sales data for food categories provides better insight into prioritizing different food categories for advocacy targets. Figure 3 shows estimated total egg content across the top 20 food categories in our results, calculated as the product of per-serving average egg use and unit sales, while Figure 4 shows the distribution of all food categories (with top and bottom categories only labeled). These results help us determine more appropriate advocacy targets; for example, the prepared subs seem to use fewer eggs per serving than milk products (the estimation results of which are driven by eggnog products), but the much higher average unit sales of prepared subs and sandwiches may make it a better candidate for egg replacement in terms of overall market impact. As discussed in the Robustness checks section, these results are not sensitive to using mean versus median calculations.

Figure 3: Total egg use in top 20 categories

Note: Top categories in terms of total egg use. Category names are taken directly from the data and do not have the same editorial formatting applied as the tables in this report.

Figure 4: Total egg use by food category, all categories

Note: Category labels are removed for clarity, except for the top and bottom categories. Category names are taken directly from the data and do not have the same editorial formatting applied as the tables in this report.

The percentage of egg-containing products in a category provides further insight into category concentration, and this statistic may be useful for corporate outreach advocacy strategies. Figure 5 shows the percentage of egg-containing products per category alongside unit sales. In particular, this plot might help to identify food categories in which food production may be more concentrated, as well as higher in sales and egg use.

Figure 5: Percent egg-containing products, total products, and sales by food category

Notes: Bubble size represents average unit sales (averaged across all products in a category). Category names are taken directly from the data and do not have the same editorial formatting applied as the tables in this report.

Observations of main egg-using products

• Milk products illustrate the importance of considering overall market size when choosing replacement priorities. Egg-containing milk products, like eggnog, appear to be impactful advocacy targets with high per-serving egg use. However, the total sales of the milk is not even in the top 20 categories by total egg use, compared to categories with similar per-serving egg use that have higher total sales and may be more effective targets.
• Similarly, pancakes, waffles, and crepes use about half an egg more per serving on average than prepared subs and sandwiches, but prepared subs and sandwiches represent a much larger market by sales volume, making it potentially a higher-impact target for egg replacement advocacy.
• Some categories show extreme outliers in egg content (omitted from the plots above), suggesting either 1) data or estimation errors are creating biased point estimates, or 2) that within certain categories, there may be specific product formulations that are particularly egg-intensive and worth targeted approaches. Before proceeding as if categories like prepared subs and sandwiches and snack, granola, and energy bars are priorities for replacement, a more targeted investigation of actual egg use would be warranted.

Egg function in food product categories

Egg function is classified using AI and expert domain knowledge. Here, we report select category-level results from the AI classification and validation of egg function within food categories. The full product-level results can be found in the supplementary file egg_function > results > egg_food_2023_joined_functions.csv.

Table 3 displays the egg functions of the top 20 categories by number of products. The full function classification results can be found in the supplementary file egg_function > results > category_summary.csv.

Table 3: Egg function classification results via AI, top 20 categories by number of products

RSSL results support our intermediate AI analysis by adding nuance to the egg functionality classification. Table 4 displays their full analysis based on their expert domain knowledge. The results denote the main form of egg used and the likely primary and secondary functions of the categories. Comments provide insight into likely market dynamics and potential replacement products for the egg functions, which we incorporate into our recommendations where appropriate.

Table 4: RSSL expert analysis results of egg function by food categories

To further incorporate the RSSL results into our overall analysis, we modify the original replaceability algorithm to create a weighted average number of functions based on primary and secondary egg functionality. This modification necessarily increased the number of functions over all categories, which resulted in all categories being classified as more “difficult” to replace than our intermediate AI results. The final results place all food categories into Priorities 3, 5, 6, and 8, although the food category order is qualitatively similar. Therefore, our final recommendations are largely the same, with additional nuance.

Observations on egg function within the data

• The most prevalent egg functions across food categories are moisture/fat contribution, binding, and structure, each appearing in a greater number of product categories than other functions (Table 3 and Figure 6 below).
• Many categories rely on eggs for multiple functions simultaneously, with certain categories like cakes requiring eggs to serve seven different functions, making replacement more technically challenging.
• Notably, none of the categories use eggs in only a single-function capacity, suggesting that “low-hanging fruit” replacement may not exist.

Product replaceability categorization

Based on the procedure defined in the Replaceability classification methods section, we categorize food product categories as lower or higher priority to target with replacement advocacy efforts, then combine these results with egg use estimations above. The full results of replacement prioritization can be found in the supplementary file egg_replaceability > results > egg_use_priority.csv.

Figure 1, reprinted here from the Key findings section above, brings together all three components of the analysis to describe the complexity of replacing egg function alongside the number of available alternatives. This summary information may allow us to identify categories that have both easily replaceable egg products and large egg use, to maximize the impact of egg replacement advocacy.

Figure 1: Egg use distribution by priority group

Notes: These estimates are low-medium confidence. Percentages are rounded to two decimal places.

Figure 6 provides further nuance to understanding the relationship between egg replacement availability and egg function prevalence. Egg functions with greater representation (especially greater representation as secondary functions, which may be easier to replace) are moisture/fat contribution, binding, and structure, while functions with the greatest number of alternatives are binding and then emulsification and moisture. These results suggest that functions like binding may be a better target for adoption outreach efforts since many products utilize the binding function of eggs (though usually one of several functions) for which there are many alternatives available, while functions like moisture or structure might be better targets for research advocacy to develop new alternatives. We did not encounter performance data for these alternatives, which suggests an opportunity to either find and centralize or conduct research to inform alternative use. Table 5 shows examples of alternatives for each egg function.

Figure 6: Egg function, category usage vs. available alternatives

Table 5: Examples of alternatives classified by target egg function

Observations on replaceability within the data

• No products are classified as Priority 1 or 2.
• In Priority 3, prepared subs, sandwiches, wraps and pasta have high sales such that, even though their replaceability complexity scores are relatively high as we have defined the criteria, the total category egg use suggests that they may be impactful replacement targets.
• Binding may represent the “sweet spot” function with both high prevalence across food categories and high number of available alternatives (Figure 6), making it an excellent target for immediate advocacy efforts.

Robustness checks

Our initial estimates of total category egg use suggested a large overestimation. For example, the grand total of one priority egg use was an estimated 6.1 billion eggs weekly, which is nearly four times greater than the US monthly egg production in December 2024 (around 8 billion eggs meant for human consumption, according to the USDA).

We investigated different specifications for egg use calculation that might provide more realistic estimation results: namely, using median versus mean values to calculate per-product egg use and other key metrics, and discounting the total category egg use calculation by the percentage of egg-containing items in each category. The former specification only produces small changes to total egg use and no qualitative changes to the overall replaceability results.

The latter specification is more theoretically compelling: we might understandably want to discount the product of unit sales and per-product egg use in order to more closely capture only total sales of egg-containing products. This more conservative estimation of total category egg use provides potentially more realistic results, and the specification does not change our replaceability categorization, only the point estimate of the number of eggs used by each category (which subsequently only changes one of the categories we recommend for deeper dives). One key limitation of this method is that we do not know the distribution of unit sales between egg-containing and egg-free items, which will likely bias the results in a direction that is unclear. Additionally, the difficult process of matching branded food categories to sales categories will lead to double counting in any specification, so we suspect that some overinflation of estimates will always remain when using these particular data. For both of these reasons, deeper dives in key categories are still highly recommended as before.

Table 6 compares select results from both specifications. Notably, the preferred (robust) specification estimates greater total egg use for whole egg-heavy categories like breakfast sandwiches and raw eggs. If we were to recommend replacement targets based only on egg use, we might focus on these categories. However, their eggs may be difficult to replace with current replacement technology. Instead, including egg function and replaceability into our recommendations leads us to other categories (e.g., subs, sandwiches, and wraps). The full egg use estimation results from both specifications can be found in Table 13 in the Appendix.

Table 6: Select results comparing initial to robust egg use calculations

Note: These estimates are low-medium confidence. Robust egg use estimates are calculated as the product of percentage of egg-containing products, per-serving egg use, and weekly unit sales, averaged over food categories. Total and percentage egg use numbers are rounded to two decimal places. Due to editorial formatting, some categories may appear to be duplicates although they are distinct in the data. These categories illustrate the importance of additional data cleaning and category investigation before taking strategic action.

Appendix

Research antecedents

Previous research describes food science functions of eggs and potential functions of egg replacements, as well as potential barriers to adopting replacements and possible interventions to surmount these barriers. IMPROVE (2023) conducted a comprehensive review of over 100 commercial egg-containing products, including bakery, mayonnaise and sauces, frozen desserts, and pasta/noodles. The authors then tested 10 commercial protein ingredients from various plant sources (soy, pea, fava bean, potato, canola, hemp) and found that while plant-based alternatives can match eggs for individual attributes (foaming, gelling, emulsifying) or even a handful of them, few possess the versatility of eggs. Some plant-based ingredients performed well in specific applications, with potato proteins demonstrating functionality closest to egg whites in terms of emulsifying, foaming, and gelling properties; however, these are only a few of the functions that egg whites serve.

The Lever Foundation’s Egg Replacement Solutions Directory (2025) collects listings for commercially available egg replacement ingredients from over 40 suppliers offering solutions across applications, including bakery, sauces, meringues, and plant-based meats, with many solutions enabling partial to complete egg replacement. While the directory entries are self-reported by the replacement manufacturers and have not been third-party tested for functionality, the directory is very valuable to egg replacement advocates as one of the only centralized sources of information on replacement products from multiple suppliers.

The RSSL report (2023) and Specht’s confidential egg-reduction technology report (2025) explored barriers to adoption of egg-replacement technologies. RSSL found that while allergen removal was the biggest driver for egg replacement, manufacturers face challenges with functionality, emulsification, organoleptic properties, and clean label considerations. Specht’s research identified additional barriers: fragmented awareness of solutions, risk aversion, lack of incentives for suppliers to promote egg replacers, and insufficient commercialization pathways for promising academic research. Both reports concluded that total egg replacement remains challenging, with reformulation of existing products being more difficult than developing new egg-free products. Enzyme technology emerged as a promising approach for egg reduction, enabling up to 30% reduction with no impact on ingredient declarations, though this represents only partial replacement. Together, these findings suggest that while technically viable egg replacement solutions exist for most applications, broader adoption will require addressing both technical and logistical barriers to egg replacement.

Methods

Our research combines three main lines of analysis: 1) estimating the rough volume of egg use across categories of manufactured food products, 2) classifying the food science functions of eggs in product categories, and 3) classifying the ease of reducing or replacing egg ingredients using alternative ingredients like plant-based proteins or starches. Our scope and methods for this MVP are largely dictated by the available data, described below. This multi-pronged approach to analysis provides a framework for strategic decision-making in plant-based food development by balancing technical feasibility of egg replacement with market impact, enabling more effective allocation of R&D resources toward the most promising egg replacement or reduction opportunities.

Data

The egg ingredient estimation procedure uses two different data products from the United States Department of Agriculture: back-of-label nutrition facts and ingredients from FoodData Central Database (using 2025 data), and market size data from Weekly Retail Sales database (using 5 months of 2023 data). The FoodData Central database records nutritional and ingredient information for individual branded food products, in which food categories are defined at the finest level relative to all of our different datasets. The Weekly Retail Sales database defines categories differently and slightly more coarsely than FoodData Central. From the Weekly Retail Sales database, we use the number of units sold weekly in a given category, averaged over the five available months in 2023.^[5]

To determine egg function in food product categories, we combine egg functionality data commissioned from RSSL, the authors of the previously commissioned manufacturers’ survey, with data from the American Egg Board to create an algorithm (using Claude AI) to assign functionality by food product category. The replaceability classification procedure uses data on egg alternatives as compiled by Lever Foundation’s egg replacement directory (2025), and egg functionality results from our own functional analysis.

Analysis methods

Egg use estimation

We investigate egg use in processed foods by applying an adapted version of the Choi and Okos (1986) estimation model, which uses total cholesterol information to estimate egg count. Our implementation of the model was developed by Claude AI from the source paper for analysis using R software. The estimation results are simply calculated^[6] from dividing total cholesterol by the amount of cholesterol in one large egg, with a minor multiplicative adjustment for moisture volume lost during cooking. For product ingredients
and nutrients
, we estimate a system of equations similar to the following model, which is based on refinements from other citing works (Miles, C. A., et al., 1997; Rahman, M. S., 2009; Sahin, S., & Sumnu, S. G., 2006; Fehaili, S. et al., 2010):

The procedure combines two key components: first, ingredient detection through text mining of ingredient lists, and then compositional estimation based on cholesterol content. For ingredient detection, we identify egg-containing products by scanning ingredient lists for egg-related terms (e.g., “egg,” “egg yolk,” “albumen”). For products lacking clear ingredient information, we use cholesterol content as a proxy indicator.

To conduct the compositional estimation, we first normalize nutrient data and serving size information to standardize units across products. For products containing eggs in the ingredient list, we calculate egg content using the ratio of cholesterol concentration in a product serving compared to a single large egg and applying a thermal correction factor to account for moisture loss. This method introduces an important limitation: the assumption that only eggs contribute to cholesterol, thereby ignoring the cholesterol contribution of other animal products. Therefore, our egg use estimate should be interpreted as the upper bound of eggs in each product. While this method does present limitations, we determined through conversations with experts that the cholesterol method is more reliable than other competing methods, namely, estimating egg use from top-down national broad category egg disappearance data, or estimating all ingredient weights using more nutritional information than just cholesterol.

Finally, we join the egg use estimates with retail sales data to provide rough market-level consumption patterns. The datasets are matched using a crosswalk table mapping branded food categories to retail categories, which we manually matched based on subjective judgment of connection between the two datasets (USDA did not provide a crosswalk between these two datasets). The retail sales dataset has fewer categories, so several branded food categories may be matched to the same retail sales category. The combined dataset allows us to identify high-impact product categories for egg consumption based on both the proportion of egg-containing products within each category and their respective sales volumes. Table 14 (posted at the document end due to length) shows the crosswalk of food product categories between the two datasets.

All analyses are conducted using R data analysis program, with code written and debugged using Claude AI. Table 7 provides an illustration of the steps of the per-serving egg ingredient estimation procedure using actual observations from the USDA FoodData Central branded foods database, and reference values from the USDA. Note that this method assumes that there is one serving per unit of product due to data limitations. Only ~300,000 out of the 1.8 million total branded food products contain information about overall package size, which could be used to calculate servings per package more precisely. The single-serving assumption proves to be the least harmful because even when available, package size data is often incorrectly recorded, leading to greater errors in egg use calculation than otherwise.

Table 7: Illustration of the cholesterol-based per-serving egg use estimation procedure

Note: The table shows estimates from two example products that will then be used alongside other estimates in the same food category to calculate average large egg use per category. Numbers are rounded to two decimal places.

Source: USDA FoodData Central branded products, USDA reference data

We validate the market size estimation methods and results through expert review. We have low confidence in the results from this estimation model, due to the limitations listed in Table 8 below, especially the assumption of cholesterol from eggs only and the data parsing issues.

Table 8: Limitations of the egg use estimation procedure and their potential impacts

Function classification

We commissioned Reading Scientific Services Ltd. (RSSL) to classify egg function in food product categories based on their food science expertise. To provide a baseline for their analysis, we constructed our own classification method. RSSL researchers conducted their analysis using our categories as we defined in our egg use and egg functionality methods.

Our analysis determines egg functionality in food products using a two-stage algorithmic approach that uses food category information to identify likely egg functions. First, we use Claude AI to develop a comprehensive egg function classification algorithm that maps food categories to probable egg functions based on a USDA database of egg functionality profiles. This algorithm assigns one or more of 20 potential egg functions (e.g., binding, emulsification, aeration/foaming/structure) to food products based on their branded food category. The algorithm then creates a detailed mapping system between food categories in our data and over 40 food category types (and the associated egg functionalities for each type), as determined by American Egg Board documentation. Table 9 illustrates the function classification procedure on two example functions.

Table 9: Function classification criteria for two example egg functions

To validate our classifications, we compare our function assignments against a smaller reference dataset from the IMPROVE (2023) report, using both exact and fuzzy matching techniques. For food categories without direct matches, we employed manual mappings based on domain knowledge of food categories. The validation demonstrated the reliability of the AI algorithmic function classification, as determined by manually evaluating the matches of a random subset of the results. The commissioned function classifications from RSSL provide additional validation for our AI-generated results.

RSSL researchers started from our previously defined food categories and assigned primary and secondary egg functionality based on their product expertise. They also provided information on the likely type of egg used (for example, whole, yolk, or egg white) and comments on possible replacement products or other market characteristics relevant to egg replacement. RSSL expertise comes from their extensive food science and product development research for Mondelez International, the parent company of RSSL, and other major food manufacturers.

We place medium confidence in the functionality results generated by an AI-created algorithm, because the procedure follows a set of systematic rules based on good underlying data; and high confidence in the results from RSSL. Table 10 details the limitations and possible directions/impacts of potential bias from these issues. In particular, egg functionality is well studied, with an in-depth understanding of egg behavior in many different types of food products. The complexity of the interaction between functions is perhaps a reason to lower our confidence in the results, since the interactions may have a superadditive effect that is difficult to define as a single function, but for that reason, we settle on medium.

Table 10: Limitations of the egg function classification and their potential impacts

Replaceability classification

In this MVP, we define replaceability in terms of the two major components of technical feasibility: how many egg functions need to be replaced, and how many potential alternatives are already available. We asked Claude AI to develop an egg replacement prioritization framework that analyzes these two factors together for each food category. The prioritization procedure assigns a replacement complexity score to each food category that increases in the number of potential alternatives, but decreases in the number of egg functions.

To assign these scores, we use three primary data inputs: counts of alternatives for each food category (based on Lever Foundation’s egg alternatives directory), category-function mappings (from the previous egg function analysis), and egg functionality by food category (from RSSL). We standardize egg function names and normalize the category-function relationship across both datasets. For each food category, we calculate key metrics including primary and secondary egg function counts, total alternatives, weighted number of egg functions with and without alternatives (where secondary functions are weighted as 0.5 and primary functions as 1). These metrics are combined into a replacement complexity score, defined as the weighted average of total egg functions discounted by the proportion of egg functions without alternatives.^[8] We use the number of alternatives as part of the proxy for replacement complexity with the logic that more solutions should be positively correlated with at least one already available solution that, with further investigation, will prove adequate in terms of functionality, price, and other important replacement factors. This proxy measure may somewhat ameliorate the limitations of our model with respect to these factors. The key metrics are used to classify food categories into a priority matrix based on complexity (low: complexity score <= 0.75; medium: 2-3 functions; high: 4+ functions) and alternative availability (low: <5 per function; medium: 5-10 per function; high: >10 per function). Table 11 illustrates the criteria for replacement prioritization described above.

Table 11: Criteria for replaceability prioritization

Note: These criteria are based primarily on weighted average of primary and secondary egg functions and number of alternatives per function. Future research should investigate other aspects of replacement feasibility.

We have not yet validated these results. Our validation plan was to rely on food expert validation of a small sample of the results. However, consultation with RSSL indicated that egg replacement functionality is difficult to classify without third-party testing, and as such, classifying replacement functionality would fall outside the scope of this MVP. We will leave validation for future projects.

We have low confidence in the overall replaceability prioritization. Table 12 summarizes these limitations and their potential impact. First, there are several known factors that are not included in our replaceability scores: namely, economic factors like cost of replacements and scalability, and consumer acceptance factors like taste, desire for clean labels, and cultural tradition. Further, the function of eggs in food products is complex, and not every potential replacement may be suitable for a given product “off the shelf,” so to speak. Recipe reformulation may require extensive testing or combinations of egg replacements. Finally, as discussed above, egg replacement functionality is hard to classify without third-party testing, so our egg-and-replacement matching procedure should only be considered a guideline for future replacement product testing and egg replaceability classification.

Table 12: Limitations of the egg replacement classification and their potential impacts

Full results from robustness check of egg use estimation

 
Table 13: Full results, comparing original and robust egg use specifications